UK: +44 (0)1223 264428
USA: +1 (650) 798 5134
by Jason Mashinchi
I wonder how many athletes in the Winter Olympics had used artificial intelligence (AI) to improve their performance. AI is when a computer is programmed to mimic human cognitive functions such as learning and problem solving. With wearables like Fitbit abounding in the sports market, AI is beginning to play a key role in allowing these devices to provide a ‘quantified self’ – useful statistics derived from performance measurements that allow athletes to improve their training, and therefore their performance.
Let’s take tennis as an example – often athletes will need a coach to give them direction on how to improve their form, and to spot mistakes and inefficiencies while watching a game. Coaches typically set targets and goals by making judgements based on what they can see. Wearables can provide insightful data on previously undetectable metrics to coaches, through readings such as heart rate and movement. Wearables contain electronic sensors, such as accelerometers or gyroscopes, which provide continuous movement information that shows variations when different events occur. For example, a wrist-worn device will read different movement patterns for a forehand shot, compared with a backhand equivalent.
Readings from wearables will be affected by many factors, such as the way the person is holding a racket, what serve or shot they’re doing or even how the device is being worn. That’s just a fraction of the many things that could change, and these variables all need to be considered to find out how the measurements match up to events.
In conventional approaches, computers use signal processing to find events in the incoming readings. The computer software needs to be programmed with various filters and transformations to apply to these readings, and these are used to detect specific, consistent events that may be occurring. However, with sporting data, it’s hard to predict what each event looks like because events rarely involve the exact same movement as there are so many factors that could change. This reduces the accuracy of the device, which isn’t ideal.
Machine learning – an area of AI – can help. Machine learning allows software to learn from incoming readings and to identify factors that affect the measurements. This allows software to decide which events are occurring in given measurements and to get better with more data over time. For example, with tennis data, machine learning can detect similar movement patterns and group similar data together – in effect, it creates movement classifications by itself for events like serves, and shots (e.g. forehand, backhand, volleys). AI lets us process vast amounts of statistical data with less effort than ever – and can even identify factors affecting sports performance that are impossible for us humans to detect.
Although machine learning can be effective, it brings its own set of challenges. AI can never be 100% accurate without vast quantities of training data. We’re talking about hundreds of thousands of readings – and it’s not easy to collect this much data manually! Luckily, it’s possible to start with a smaller amount of data and gradually train the machine learning models with more data over time to improve accuracy.
By processing wearable sensor readings using AI, valuable metrics that were previously difficult to gather can be obtained. By combining this data with powerful smartphone apps, athletes can be provided with actionable insights and coaching, which in turn improves performance and helps athletes to achieve their best.
Our in-house software experts here at Cambridge Design Partnership have delivered successful implementations of AI to many of our clients. Just think what AI could do for your business…
A look at why corporates in China are increasingly focusing on open innovation.
18 July 2018
Why it’s crucial to design and engineer with safety in mind right from the start.
16 July 2018
Stay up to date with all our work and our latest news by signing up to our newsletter.